Using TFLite GPU about flutter_tflite HOT 5 CLOSED

I got GPU delegate to work by removing unnecessary import from my project (I don't remember exactly how, but I had made really silly mistake. I guess I had gradle tensorflow gpu import inside my project, not in plugin). But I couldn't run inference on tiny yolov2 because it has operations that are unsupported by tensorflow lite gpu delegate (if I recall correctly it was MAX operation).

How did you enable/add GPU delegate in your Flutter project?

from flutter_tflite.

GPU delegate is now supported in the latest version 1.1.0. Set useGpuDelegate to true in loadModel.

Thanks,
Qian

from flutter_tflite.

Hi, @shaqian
It's a great work! I also tried to apply GPU delegate by adding gpudelegate to .\flutter.pub-cache\hosted\pub.flutter-io.cn\tflite-1.0.4, and also implemented tensorflow-lite-gpu:0.0.0-nightly in dependencies.
However it doesn't seems a huge difference occurs, looking forward to your suggestions!

from flutter_tflite.

Sorry for the delay. I don't have a compatible phone to test GPU. 😂
Looking at the the error, I think the problem may be the path is wrong? Should it be "/org" rather "Lorg" ?
Failed resolution of: Lorg/tensorflow/lite/experimental/GpuDelegate;

@Cheren15 Are you able to get GPU work successfully? I saw discussions previously (if I remember correctly) the GPU version improves certain operations, if your model doesn't use those operations there won't be much difference.

from flutter_tflite.

I got GPU delegate to work by removing unnecessary import from my project (I don't remember exactly how, but I had made really silly mistake. I guess I had gradle tensorflow gpu import inside my project, not in plugin). But I couldn't run inference on tiny yolov2 because it has operations that are unsupported by tensorflow lite gpu delegate (if I recall correctly it was MAX operation).

from flutter_tflite.